Filtering spatial error from DEMs: Implications for morphological change estimation

Scour and fill estimation from digital elevation model (DEM) subtraction or differencing is an increasingly common technique in morphological and sediment transport investigations. The technique is commonly used to estimate scour and fill volumes and to produce scour and fill maps that provide process-based information to geomorphologists. Accounting for sources of uncertainty within the DEM is of critical importance. DEM error is spatially variable and has a tendency to be greater at breaks of slope such as bar and bank edges. In the past however, this has been achieved using a uniform error metric across the DEM, resulting in over-conservative estimates of error. In turn this has led to over-conservative scour and fill volumes, and incorrect process interpretation. This paper applies a new approach that permits assessment of spatially distributed error across a DEM. The method is tested on a sequence of field surveys of the gravel-bed River Nent, Cumbria, UK. The results demonstrate some dramatic differences: application of conventional techniques that account for mean error across a DEM led to a 15 and 31% underestimation in scour and fill volumes, respectively, between July and October 1998, whilst for the October 1998–June 1999 subtraction 31 and 13% of scour and fill were underestimated respectively. Use of a uniform error across a surface captures less change in comparison to a spatially distributed approach. Furthermore, the changes captured using a uniform error are biased toward areas of the channel that have more local topographic variability such as bar and bank edges. In contrast the use of a spatially distributed approach provides information on change from flatter surfaces such as bar tops that would otherwise be missed. This study also demonstrates that estimates of morphological change can be misleading in the absence of an error filter. Where the raw survey data is available, it is recommended that sediment budgeting studies take account of the spatial variability of error in each DEM involved in the subtraction.